> ## Content Index > Fetch the complete content index at: https://www.implicator.ai/llms.txt > Use this file to discover other available public pages before exploring further. # LLM Meter — Week of July 16, 2026 - URL: https://www.implicator.ai/llm-meter-week-of-july-16-2026/ - Published: 2026-07-16T08:23:04.000Z - Updated: 2026-07-16T08:23:04.000Z - Author: Marcus Schuler - Tags: #llm-meter \---CHATGPT--- score: 90 trend: up change: +1 \+ GPT-5.6 reached general availability July 9 after the government gate lifted, with day-one Amazon Bedrock listing and Microsoft 365 Copilot switching to it as the default model \+ ChatGPT Work launched the same day, an agent that runs multi-hour office tasks across Slack, Salesforce and Google Drive for 9M-plus paying business seats \+ Sol leads the agentic benchmarks with 88.8% on Terminal-Bench 2.1 and a record 53.6 on Agents' Last Exam, and usage caps on Codex and Work were temporarily lifted \- Apple sued OpenAI July 10 for trade-secret theft and seeks an injunction on its hardware line, while the NYT asked the court to sanction OpenAI for allegedly deleting chat-log evidence \- Safety chief Johannes Heidecke became the sixth safety leader to exit in two years, and Ramp now shows OpenAI trailing Anthropic in tracked enterprise spend \---CLAUDE--- score: 84 trend: up change: +4 \+ Fable 5 returned to worldwide general availability July 1 with a new classifier blocking the flagged jailbreak in over 99% of attempts, ending the 19-day export suspension that had the flagship dark \+ IPO machinery engaged: Goldman, Morgan Stanley and JPMorgan are scheduling investor meetings for a listing as soon as October at a $965B valuation, with Yipit estimating ARR near $69B \+ Ramp data shows Anthropic passing OpenAI in tracked US enterprise adoption, 34.4% vs 32.3%, while Fable 5 holds the coding crown at 80% on SWE-Bench Pro against GPT-5.6 Sol's 64.6% \+ Ode with Anthropic, a Blackstone-backed enterprise AI-services firm, launched July 15 to push midsize deployments as Claude Cowork hit web and mobile with 600,000-plus organizations \- Alibaba banned Claude Code July 10 after researchers found hidden China-fingerprinting code in the client, and a roughly 18-hour July 7 outage hit Fable 5 within a week of its relaunch \---GEMINI--- score: 81 trend: down change: -4 \+ Gemini Enterprise claims over 8M paid seats across 2,800-plus companies with paid usage up 40% quarter over quarter, and Cognizant expanded its partnership July 7 \+ Consumer momentum holds at 27.4% web share and 118M daily users in June, the largest daily-user gain of any tracked AI app \- Gemini 3.5 Pro missed general availability a fifth straight month after DeepMind reportedly scrapped the base model for a rebuild; the July 17 target remains unconfirmed by Google \- The deployability advantage evaporated as Fable 5 and GPT-5.6 both returned to full availability while Gemini's flagship stays in preview \- The EU's top court made the €4.1B Android fine final July 2, opening follow-on damages suits, with a record DMA fine over Search self-preferencing due by July 27 \---MISTRAL--- score: 74 trend: down change: -2 \+ An open-weight frontier model entered early access with government and industry partners, Mistral's first credible answer to the top-end gap, with broad release promised this summer \+ EU AI Act enforcement went live July 10 with GPAI obligations from August 2, strengthening the compliance-ready sovereign pitch, and CI&T signed on as preferred LATAM partner \+ Release cadence held with Leanstral 1.5 for formal verification, the Robostral Navigate robotics model, and enterprise prompt governance in AI Studio \- The reported €3B raise at a roughly €20B valuation is still not closed five weeks after Bloomberg first reported the talks \- The sovereignty tailwind weakened as both US flagships returned to unrestricted worldwide availability during the window \---GROK--- score: 30 trend: down change: -4 \+ Grok 4.5, the 1.5T-parameter Cursor-trained model, shipped broadly July 8 at $2/$6 per million tokens to undercut Claude, with the signed $60B Cursor acquisition set to close in Q3 \- Researchers showed Grok Build's CLI uploaded entire Git repositories including committed secrets to xAI cloud storage while the privacy toggle did nothing, a disqualifying finding for enterprise code work \- Independent benchmarks place Grok 4.5 fourth on the Artificial Analysis Intelligence Index, and Musk conceded Fable "is definitely better than Grok 4.5" \- Reporting surfaced 59 unpermitted gas turbines at Colossus 2, roughly double what xAI acknowledged, with the DOJ intervening on national-security grounds \- Grok 4.5 launched without EU availability and is absent from Azure AI Foundry while the EU's DSA investigation stays open \---DEEPSEEK--- score: 24 trend: up change: +4 \+ DeepSeek became the top model line on OpenRouter at 17.6% of weekly tokens, more than Google and OpenAI combined on that gateway, and reached about 23% of tokens on Vercel's AI Gateway \+ IPO preparation began for a 2027 mainland listing alongside talks for a $1.5B round at a $71B valuation, up roughly 37% in six weeks \+ V4 general availability was announced for mid-July with a 1M-token context window and peak/off-peak pricing that preserves the cost floor, and an in-house inference chip is in development \- Beijing is weighing curbs on overseas access to advanced Chinese models, a second political chokepoint that directly threatens the router-driven adoption earning this week's points \- State-fund board control and 17-plus US state bans plus federal restrictions keep DeepSeek off-limits for regulated Western buyers